Spaces:
Running
on
A10G
Running
on
A10G
File size: 24,921 Bytes
b582ef0 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 |
from typing import Dict
import numpy as np
from omegaconf import DictConfig, ListConfig
import torch
from torch.utils.data import Dataset
from pathlib import Path
import json
from PIL import Image
from torchvision import transforms
from einops import rearrange
from typing import Literal, Tuple, Optional, Any
import cv2
import random
import json
import os, sys
import math
import PIL.Image
from .normal_utils import trans_normal, normal2img, img2normal
import pdb
def shift_list(lst, n):
length = len(lst)
n = n % length # Ensure n is within the range of the list length
return lst[-n:] + lst[:-n]
class ObjaverseDataset(Dataset):
def __init__(self,
root_dir: str,
num_views: int,
bg_color: Any,
img_wh: Tuple[int, int],
object_list: str,
groups_num: int=1,
validation: bool = False,
random_views: bool = False,
num_validation_samples: int = 64,
num_samples: Optional[int] = None,
invalid_list: Optional[str] = None,
trans_norm_system: bool = True, # if True, transform all normals map into the cam system of front view
augment_data: bool = False,
read_normal: bool = True,
read_color: bool = False,
read_depth: bool = False,
mix_color_normal: bool = False,
random_view_and_domain: bool = False
) -> None:
"""Create a dataset from a folder of images.
If you pass in a root directory it will be searched for images
ending in ext (ext can be a list)
"""
self.root_dir = Path(root_dir)
self.num_views = num_views
self.bg_color = bg_color
self.validation = validation
self.num_samples = num_samples
self.trans_norm_system = trans_norm_system
self.augment_data = augment_data
self.invalid_list = invalid_list
self.groups_num = groups_num
print("augment data: ", self.augment_data)
self.img_wh = img_wh
self.read_normal = read_normal
self.read_color = read_color
self.read_depth = read_depth
self.mix_color_normal = mix_color_normal # mix load color and normal maps
self.random_view_and_domain = random_view_and_domain # load normal or rgb of a single view
self.random_views = random_views
if not self.random_views:
if self.num_views == 4:
self.view_types = ['front', 'right', 'back', 'left']
elif self.num_views == 5:
self.view_types = ['front', 'front_right', 'right', 'back', 'left']
elif self.num_views == 6 or self.num_views==1:
self.view_types = ['front', 'front_right', 'right', 'back', 'left', 'front_left']
else:
self.view_types = ['front', 'front_right', 'right', 'back', 'left', 'front_left']
self.fix_cam_pose_dir = "./mvdiffusion/data/fixed_poses/nine_views"
self.fix_cam_poses = self.load_fixed_poses() # world2cam matrix
if object_list is not None:
with open(object_list) as f:
self.objects = json.load(f)
self.objects = [os.path.basename(o).replace(".glb", "") for o in self.objects]
else:
self.objects = os.listdir(self.root_dir)
self.objects = sorted(self.objects)
if self.invalid_list is not None:
with open(self.invalid_list) as f:
self.invalid_objects = json.load(f)
self.invalid_objects = [os.path.basename(o).replace(".glb", "") for o in self.invalid_objects]
else:
self.invalid_objects = []
self.all_objects = set(self.objects) - (set(self.invalid_objects) & set(self.objects))
self.all_objects = list(self.all_objects)
if not validation:
self.all_objects = self.all_objects[:-num_validation_samples]
else:
self.all_objects = self.all_objects[-num_validation_samples:]
if num_samples is not None:
self.all_objects = self.all_objects[:num_samples]
print("loading ", len(self.all_objects), " objects in the dataset")
if self.mix_color_normal:
self.backup_data = self.__getitem_mix__(0, "9438abf986c7453a9f4df7c34aa2e65b")
elif self.random_view_and_domain:
self.backup_data = self.__getitem_random_viewanddomain__(0, "9438abf986c7453a9f4df7c34aa2e65b")
else:
self.backup_data = self.__getitem_norm__(0, "9438abf986c7453a9f4df7c34aa2e65b") # "66b2134b7e3645b29d7c349645291f78")
def __len__(self):
return len(self.objects)*self.total_view
def load_fixed_poses(self):
poses = {}
for face in self.view_types:
RT = np.loadtxt(os.path.join(self.fix_cam_pose_dir,'%03d_%s_RT.txt'%(0, face)))
poses[face] = RT
return poses
def cartesian_to_spherical(self, xyz):
ptsnew = np.hstack((xyz, np.zeros(xyz.shape)))
xy = xyz[:,0]**2 + xyz[:,1]**2
z = np.sqrt(xy + xyz[:,2]**2)
theta = np.arctan2(np.sqrt(xy), xyz[:,2]) # for elevation angle defined from Z-axis down
#ptsnew[:,4] = np.arctan2(xyz[:,2], np.sqrt(xy)) # for elevation angle defined from XY-plane up
azimuth = np.arctan2(xyz[:,1], xyz[:,0])
return np.array([theta, azimuth, z])
def get_T(self, target_RT, cond_RT):
R, T = target_RT[:3, :3], target_RT[:, -1]
T_target = -R.T @ T # change to cam2world
R, T = cond_RT[:3, :3], cond_RT[:, -1]
T_cond = -R.T @ T
theta_cond, azimuth_cond, z_cond = self.cartesian_to_spherical(T_cond[None, :])
theta_target, azimuth_target, z_target = self.cartesian_to_spherical(T_target[None, :])
d_theta = theta_target - theta_cond
d_azimuth = (azimuth_target - azimuth_cond) % (2 * math.pi)
d_z = z_target - z_cond
# d_T = torch.tensor([d_theta.item(), math.sin(d_azimuth.item()), math.cos(d_azimuth.item()), d_z.item()])
return d_theta, d_azimuth
def get_bg_color(self):
if self.bg_color == 'white':
bg_color = np.array([1., 1., 1.], dtype=np.float32)
elif self.bg_color == 'black':
bg_color = np.array([0., 0., 0.], dtype=np.float32)
elif self.bg_color == 'gray':
bg_color = np.array([0.5, 0.5, 0.5], dtype=np.float32)
elif self.bg_color == 'random':
bg_color = np.random.rand(3)
elif self.bg_color == 'three_choices':
white = np.array([1., 1., 1.], dtype=np.float32)
black = np.array([0., 0., 0.], dtype=np.float32)
gray = np.array([0.5, 0.5, 0.5], dtype=np.float32)
bg_color = random.choice([white, black, gray])
elif isinstance(self.bg_color, float):
bg_color = np.array([self.bg_color] * 3, dtype=np.float32)
else:
raise NotImplementedError
return bg_color
def load_mask(self, img_path, return_type='np'):
# not using cv2 as may load in uint16 format
# img = cv2.imread(img_path, cv2.IMREAD_UNCHANGED) # [0, 255]
# img = cv2.resize(img, self.img_wh, interpolation=cv2.INTER_CUBIC)
# pil always returns uint8
img = np.array(Image.open(img_path).resize(self.img_wh))
img = np.float32(img > 0)
assert len(np.shape(img)) == 2
if return_type == "np":
pass
elif return_type == "pt":
img = torch.from_numpy(img)
else:
raise NotImplementedError
return img
def load_image(self, img_path, bg_color, alpha, return_type='np'):
# not using cv2 as may load in uint16 format
# img = cv2.imread(img_path, cv2.IMREAD_UNCHANGED) # [0, 255]
# img = cv2.resize(img, self.img_wh, interpolation=cv2.INTER_CUBIC)
# pil always returns uint8
img = np.array(Image.open(img_path).resize(self.img_wh))
img = img.astype(np.float32) / 255. # [0, 1]
assert img.shape[-1] == 3 # RGB
if alpha.shape[-1] != 1:
alpha = alpha[:, :, None]
img = img[...,:3] * alpha + bg_color * (1 - alpha)
if return_type == "np":
pass
elif return_type == "pt":
img = torch.from_numpy(img)
else:
raise NotImplementedError
return img
def load_depth(self, img_path, bg_color, alpha, return_type='np'):
# not using cv2 as may load in uint16 format
# img = cv2.imread(img_path, cv2.IMREAD_UNCHANGED) # [0, 255]
# img = cv2.resize(img, self.img_wh, interpolation=cv2.INTER_CUBIC)
# pil always returns uint8
img = np.array(Image.open(img_path).resize(self.img_wh))
img = img.astype(np.float32) / 65535. # [0, 1]
img[img > 0.4] = 0
img = img / 0.4
assert img.ndim == 2 # depth
img = np.stack([img]*3, axis=-1)
if alpha.shape[-1] != 1:
alpha = alpha[:, :, None]
# print(np.max(img[:, :, 0]))
img = img[...,:3] * alpha + bg_color * (1 - alpha)
if return_type == "np":
pass
elif return_type == "pt":
img = torch.from_numpy(img)
else:
raise NotImplementedError
return img
def load_normal(self, img_path, bg_color, alpha, RT_w2c=None, RT_w2c_cond=None, return_type='np'):
# not using cv2 as may load in uint16 format
# img = cv2.imread(img_path, cv2.IMREAD_UNCHANGED) # [0, 255]
# img = cv2.resize(img, self.img_wh, interpolation=cv2.INTER_CUBIC)
# pil always returns uint8
normal = np.array(Image.open(img_path).resize(self.img_wh))
assert normal.shape[-1] == 3 # RGB
normal = trans_normal(img2normal(normal), RT_w2c, RT_w2c_cond)
img = (normal*0.5 + 0.5).astype(np.float32) # [0, 1]
if alpha.shape[-1] != 1:
alpha = alpha[:, :, None]
img = img[...,:3] * alpha + bg_color * (1 - alpha)
if return_type == "np":
pass
elif return_type == "pt":
img = torch.from_numpy(img)
else:
raise NotImplementedError
return img
def __len__(self):
return len(self.all_objects)
def __getitem_mix__(self, index, debug_object=None):
if debug_object is not None:
object_name = debug_object #
set_idx = random.sample(range(0, self.groups_num), 1)[0] # without replacement
else:
object_name = self.all_objects[index%len(self.all_objects)]
set_idx = 0
if self.augment_data:
cond_view = random.sample(self.view_types, k=1)[0]
else:
cond_view = 'front'
if random.random() < 0.5:
read_color, read_normal, read_depth = True, False, False
else:
read_color, read_normal, read_depth = False, True, True
read_normal = read_normal & self.read_normal
read_depth = read_depth & self.read_depth
assert (read_color and (read_normal or read_depth)) is False
view_types = self.view_types
cond_w2c = self.fix_cam_poses[cond_view]
tgt_w2cs = [self.fix_cam_poses[view] for view in view_types]
elevations = []
azimuths = []
# get the bg color
bg_color = self.get_bg_color()
cond_alpha = self.load_mask(os.path.join(self.root_dir, object_name[:3], object_name, "mask_%03d_%s.png" % (set_idx, cond_view)), return_type='np')
img_tensors_in = [
self.load_image(os.path.join(self.root_dir, object_name[:3], object_name, "rgb_%03d_%s.png" % (set_idx, cond_view)), bg_color, cond_alpha, return_type='pt').permute(2, 0, 1)
] * self.num_views
img_tensors_out = []
for view, tgt_w2c in zip(view_types, tgt_w2cs):
img_path = os.path.join(self.root_dir, object_name[:3], object_name, "rgb_%03d_%s.png" % (set_idx, view))
mask_path = os.path.join(self.root_dir, object_name[:3], object_name, "mask_%03d_%s.png" % (set_idx, view))
normal_path = os.path.join(self.root_dir, object_name[:3], object_name, "normals_%03d_%s.png" % (set_idx, view))
depth_path = os.path.join(self.root_dir, object_name[:3], object_name, "depth_%03d_%s.png" % (set_idx, view))
alpha = self.load_mask(mask_path, return_type='np')
if read_color:
img_tensor = self.load_image(img_path, bg_color, alpha, return_type="pt")
img_tensor = img_tensor.permute(2, 0, 1)
img_tensors_out.append(img_tensor)
if read_normal:
normal_tensor = self.load_normal(normal_path, bg_color, alpha, RT_w2c=tgt_w2c, RT_w2c_cond=cond_w2c, return_type="pt").permute(2, 0, 1)
img_tensors_out.append(normal_tensor)
if read_depth:
depth_tensor = self.load_depth(depth_path, bg_color, alpha, return_type="pt").permute(2, 0, 1)
img_tensors_out.append(depth_tensor)
# evelations, azimuths
elevation, azimuth = self.get_T(tgt_w2c, cond_w2c)
elevations.append(elevation)
azimuths.append(azimuth)
img_tensors_in = torch.stack(img_tensors_in, dim=0).float() # (Nv, 3, H, W)
img_tensors_out = torch.stack(img_tensors_out, dim=0).float() # (Nv, 3, H, W)
elevations = torch.as_tensor(elevations).float().squeeze(1)
azimuths = torch.as_tensor(azimuths).float().squeeze(1)
elevations_cond = torch.as_tensor([0] * self.num_views).float() # fixed only use 4 views to train
camera_embeddings = torch.stack([elevations_cond, elevations, azimuths], dim=-1) # (Nv, 3)
normal_class = torch.tensor([1, 0]).float()
normal_task_embeddings = torch.stack([normal_class]*self.num_views, dim=0) # (Nv, 2)
color_class = torch.tensor([0, 1]).float()
color_task_embeddings = torch.stack([color_class]*self.num_views, dim=0) # (Nv, 2)
if read_normal or read_depth:
task_embeddings = normal_task_embeddings
if read_color:
task_embeddings = color_task_embeddings
return {
'elevations_cond': elevations_cond,
'elevations_cond_deg': torch.rad2deg(elevations_cond),
'elevations': elevations,
'azimuths': azimuths,
'elevations_deg': torch.rad2deg(elevations),
'azimuths_deg': torch.rad2deg(azimuths),
'imgs_in': img_tensors_in,
'imgs_out': img_tensors_out,
'camera_embeddings': camera_embeddings,
'task_embeddings': task_embeddings
}
def __getitem_random_viewanddomain__(self, index, debug_object=None):
if debug_object is not None:
object_name = debug_object #
set_idx = random.sample(range(0, self.groups_num), 1)[0] # without replacement
else:
object_name = self.all_objects[index%len(self.all_objects)]
set_idx = 0
if self.augment_data:
cond_view = random.sample(self.view_types, k=1)[0]
else:
cond_view = 'front'
if random.random() < 0.5:
read_color, read_normal, read_depth = True, False, False
else:
read_color, read_normal, read_depth = False, True, True
read_normal = read_normal & self.read_normal
read_depth = read_depth & self.read_depth
assert (read_color and (read_normal or read_depth)) is False
view_types = self.view_types
cond_w2c = self.fix_cam_poses[cond_view]
tgt_w2cs = [self.fix_cam_poses[view] for view in view_types]
elevations = []
azimuths = []
# get the bg color
bg_color = self.get_bg_color()
cond_alpha = self.load_mask(os.path.join(self.root_dir, object_name[:3], object_name, "mask_%03d_%s.png" % (set_idx, cond_view)), return_type='np')
img_tensors_in = [
self.load_image(os.path.join(self.root_dir, object_name[:3], object_name, "rgb_%03d_%s.png" % (set_idx, cond_view)), bg_color, cond_alpha, return_type='pt').permute(2, 0, 1)
] * self.num_views
img_tensors_out = []
random_viewidx = random.randint(0, len(view_types)-1)
for view, tgt_w2c in zip([view_types[random_viewidx]], [tgt_w2cs[random_viewidx]]):
img_path = os.path.join(self.root_dir, object_name[:3], object_name, "rgb_%03d_%s.png" % (set_idx, view))
mask_path = os.path.join(self.root_dir, object_name[:3], object_name, "mask_%03d_%s.png" % (set_idx, view))
normal_path = os.path.join(self.root_dir, object_name[:3], object_name, "normals_%03d_%s.png" % (set_idx, view))
depth_path = os.path.join(self.root_dir, object_name[:3], object_name, "depth_%03d_%s.png" % (set_idx, view))
alpha = self.load_mask(mask_path, return_type='np')
if read_color:
img_tensor = self.load_image(img_path, bg_color, alpha, return_type="pt")
img_tensor = img_tensor.permute(2, 0, 1)
img_tensors_out.append(img_tensor)
if read_normal:
normal_tensor = self.load_normal(normal_path, bg_color, alpha, RT_w2c=tgt_w2c, RT_w2c_cond=cond_w2c, return_type="pt").permute(2, 0, 1)
img_tensors_out.append(normal_tensor)
if read_depth:
depth_tensor = self.load_depth(depth_path, bg_color, alpha, return_type="pt").permute(2, 0, 1)
img_tensors_out.append(depth_tensor)
# evelations, azimuths
elevation, azimuth = self.get_T(tgt_w2c, cond_w2c)
elevations.append(elevation)
azimuths.append(azimuth)
img_tensors_in = torch.stack(img_tensors_in, dim=0).float() # (Nv, 3, H, W)
img_tensors_out = torch.stack(img_tensors_out, dim=0).float() # (Nv, 3, H, W)
elevations = torch.as_tensor(elevations).float().squeeze(1)
azimuths = torch.as_tensor(azimuths).float().squeeze(1)
elevations_cond = torch.as_tensor([0] * self.num_views).float() # fixed only use 4 views to train
camera_embeddings = torch.stack([elevations_cond, elevations, azimuths], dim=-1) # (Nv, 3)
normal_class = torch.tensor([1, 0]).float()
normal_task_embeddings = torch.stack([normal_class]*self.num_views, dim=0) # (Nv, 2)
color_class = torch.tensor([0, 1]).float()
color_task_embeddings = torch.stack([color_class]*self.num_views, dim=0) # (Nv, 2)
if read_normal or read_depth:
task_embeddings = normal_task_embeddings
if read_color:
task_embeddings = color_task_embeddings
return {
'elevations_cond': elevations_cond,
'elevations_cond_deg': torch.rad2deg(elevations_cond),
'elevations': elevations,
'azimuths': azimuths,
'elevations_deg': torch.rad2deg(elevations),
'azimuths_deg': torch.rad2deg(azimuths),
'imgs_in': img_tensors_in,
'imgs_out': img_tensors_out,
'camera_embeddings': camera_embeddings,
'task_embeddings': task_embeddings
}
def __getitem_norm__(self, index, debug_object=None):
if debug_object is not None:
object_name = debug_object #
set_idx = random.sample(range(0, self.groups_num), 1)[0] # without replacement
else:
object_name = self.all_objects[index%len(self.all_objects)]
set_idx = 0
if self.augment_data:
cond_view = random.sample(self.view_types, k=1)[0]
else:
cond_view = 'front'
# if self.random_views:
# view_types = ['front']+random.sample(self.view_types[1:], 3)
# else:
# view_types = self.view_types
view_types = self.view_types
cond_w2c = self.fix_cam_poses[cond_view]
tgt_w2cs = [self.fix_cam_poses[view] for view in view_types]
elevations = []
azimuths = []
# get the bg color
bg_color = self.get_bg_color()
cond_alpha = self.load_mask(os.path.join(self.root_dir, object_name[:3], object_name, "mask_%03d_%s.png" % (set_idx, cond_view)), return_type='np')
img_tensors_in = [
self.load_image(os.path.join(self.root_dir, object_name[:3], object_name, "rgb_%03d_%s.png" % (set_idx, cond_view)), bg_color, cond_alpha, return_type='pt').permute(2, 0, 1)
] * self.num_views
img_tensors_out = []
normal_tensors_out = []
for view, tgt_w2c in zip(view_types, tgt_w2cs):
img_path = os.path.join(self.root_dir, object_name[:3], object_name, "rgb_%03d_%s.png" % (set_idx, view))
mask_path = os.path.join(self.root_dir, object_name[:3], object_name, "mask_%03d_%s.png" % (set_idx, view))
alpha = self.load_mask(mask_path, return_type='np')
if self.read_color:
img_tensor = self.load_image(img_path, bg_color, alpha, return_type="pt")
img_tensor = img_tensor.permute(2, 0, 1)
img_tensors_out.append(img_tensor)
if self.read_normal:
normal_path = os.path.join(self.root_dir, object_name[:3], object_name, "normals_%03d_%s.png" % (set_idx, view))
normal_tensor = self.load_normal(normal_path, bg_color, alpha, RT_w2c=tgt_w2c, RT_w2c_cond=cond_w2c, return_type="pt").permute(2, 0, 1)
normal_tensors_out.append(normal_tensor)
# evelations, azimuths
elevation, azimuth = self.get_T(tgt_w2c, cond_w2c)
elevations.append(elevation)
azimuths.append(azimuth)
img_tensors_in = torch.stack(img_tensors_in, dim=0).float() # (Nv, 3, H, W)
if self.read_color:
img_tensors_out = torch.stack(img_tensors_out, dim=0).float() # (Nv, 3, H, W)
if self.read_normal:
normal_tensors_out = torch.stack(normal_tensors_out, dim=0).float() # (Nv, 3, H, W)
elevations = torch.as_tensor(elevations).float().squeeze(1)
azimuths = torch.as_tensor(azimuths).float().squeeze(1)
elevations_cond = torch.as_tensor([0] * self.num_views).float() # fixed only use 4 views to train
camera_embeddings = torch.stack([elevations_cond, elevations, azimuths], dim=-1) # (Nv, 3)
normal_class = torch.tensor([1, 0]).float()
normal_task_embeddings = torch.stack([normal_class]*self.num_views, dim=0) # (Nv, 2)
color_class = torch.tensor([0, 1]).float()
color_task_embeddings = torch.stack([color_class]*self.num_views, dim=0) # (Nv, 2)
return {
'elevations_cond': elevations_cond,
'elevations_cond_deg': torch.rad2deg(elevations_cond),
'elevations': elevations,
'azimuths': azimuths,
'elevations_deg': torch.rad2deg(elevations),
'azimuths_deg': torch.rad2deg(azimuths),
'imgs_in': img_tensors_in,
'imgs_out': img_tensors_out,
'normals_out': normal_tensors_out,
'camera_embeddings': camera_embeddings,
'normal_task_embeddings': normal_task_embeddings,
'color_task_embeddings': color_task_embeddings
}
def __getitem__(self, index):
try:
if self.mix_color_normal:
data = self.__getitem_mix__(index)
elif self.random_view_and_domain:
data = self.__getitem_random_viewanddomain__(index)
else:
data = self.__getitem_norm__(index)
return data
except:
print("load error ", self.all_objects[index%len(self.all_objects)] )
return self.backup_data
class ConcatDataset(torch.utils.data.Dataset):
def __init__(self, datasets, weights):
self.datasets = datasets
self.weights = weights
self.num_datasets = len(datasets)
def __getitem__(self, i):
chosen = random.choices(self.datasets, self.weights, k=1)[0]
return chosen[i]
def __len__(self):
return max(len(d) for d in self.datasets)
if __name__ == "__main__":
train_dataset = ObjaverseDataset(
root_dir="/ghome/l5/xxlong/.objaverse/hf-objaverse-v1/renderings",
size=(128, 128),
ext="hdf5",
default_trans=torch.zeros(3),
return_paths=False,
total_view=8,
validation=False,
object_list=None,
views_mode='fourviews'
)
data0 = train_dataset[0]
data1 = train_dataset[50]
# print(data)
|